Comparative Analysis of the Interferogram Sensitivity to Wavefront Aberrations Recorded with Plane and Cylindrical Reference Beams

The paper investigates the sensitivity of interferograms formed using the structured reference beams. The parameters of the reference beam are selected to improve the visualization of aberrations in the interferograms. A study carried out on the use of reference beams with cylindrical wavefronts in the interferograms formation to improve the aberrations recognition using a convolutional neural network. The applying of a cylindrical reference beam instead of a plane one in the interference method for recognition of wave aberrations based on neural networks with Xception architecture makes it possible to reduce the mean absolute error by more than 30%. In this work, for each type of interferogram, the model was trained for 80 epochs, which took about 1.8 hours using GeForce RTX 2070 graphics card. However, after completing this training once, we obtain a model that allows us to make forecasts in 0.055 s for every new interferogram of the same type.